Background
Most SWMM-based LID evaluations rely on fixed AMC assumptions. In monsoon climates, however, soil moisture fluctuates dramatically across dry spells and prolonged wet sequences, so static AMC introduces systematic model error.
Approach
- GIS-integrated sewer network: 2,497-conduit trunk network for Seocho-gu and Gangnam-gu
- 2-year continuous simulation: hourly rainfall + Green-Ampt IMD dynamic coupling (2019–2020)
- 4 LID types × 4 coverage steps (10–40%) × 15 scenarios
- XGBoost surrogate + SHAP for variance attribution
Key Results
| Metric | Value | |---|---| | Peak-flow shift from AMC variation alone | −5.6% ~ +6.2% | | Dynamic AMC vs AMC II baseline | −0.9% peak | | Improvement during 2020 extreme monsoon | up to +6.2% | | Annual runoff reduction at Mixed LID 40% | 39.6% | | Peak-flow attenuation at same scenario | 33.8% | | Unit efficiency at 10% coverage | up to 98.3% | | SHAP top predictor (monsoon months) | IMD (initial moisture deficit) |
Implications
- AMC alone shifts peak flow by a magnitude equivalent to 10–15% LID coverage.
- Cost-effectiveness comparisons under static AMC conflate intervention effect with model error.
- The framework offers quantitative, evidence-based guidance for LID planning in East-Asian monsoon megacities.
Related solutions
- Stormwater Risk Analysis (SWMM · Neural ODE) — core technology behind dynamic AMC coupling
- Spatial Optimization of Green Infrastructure — extends LID scenario comparison via NSGA-II